agent-friend Schema Slimmer

Open a URL in a browser and return its page text. Requires the agent-browser CLI on the server's PATH.

Retrieves text from an HTTP or HTTPS URL and converts HTML to plain text. Output is truncated at the chosen character limit, which defaults to 8,000.

Extracts readable text from an HTML string, removing tags and script and style blocks while preserving paragraph breaks.

Extracts link text and destinations from HTML. A supplied base URL resolves relative links to full URLs.

Extracts headings from HTML, including each heading's level from h1 to h6 and its text, to show the page's structure.

Extracts a page's title and meta tags from HTML, including descriptions, keywords and Open Graph metadata.

Extracts HTML tables as rows of cell text for working with tabular page content.

Extracts the text of HTML elements matching a tag name and optional attributes, such as spans with a particular class.

Retrieves recent items from an RSS or Atom feed URL, including titles, links and summaries.

Saves a feed URL under a short name for later reading.

Lists saved feed names and their URLs.

Retrieves recent items from a feed saved under a name.

Removes a saved feed subscription by name.

Find web pages through DuckDuckGo, with titles, URLs and short excerpts in the results. No API key is required.

Extract text values from every XML element with a specified tag name.

Collect the attributes of every XML element with a specified tag name.

Find the first XML element matching an XPath expression and inspect its tag, text, attributes and child tag names.

Find all XML elements matching an XPath expression, returning each element’s tag, text and attributes.

Convert XML into a nested dictionary returned as JSON. Attribute keys use @, repeated tags become lists, and text alongside child elements uses #text.

Check whether an XML string is well-formed. This checks document syntax rather than conformance to an XML schema.

Explore an XML document’s structure by listing its unique tag names and counting their occurrences.

Compares two text strings and produces a unified diff, with counts of added, removed and unchanged lines.

Reads two local files and produces a unified diff with line-change counts. Returns an error if either file cannot be read.

Highlights word changes between two strings inline, marking additions with + and removals with -, and counts both.

Measures how similar two texts are and counts added or removed characters and lines.

Ranks candidate strings by similarity to a query, helping find approximate matches or near-duplicates.

Reads a local file within the allowed base directory. By default, files larger than 32 KB are truncated with a notice.

Creates or overwrites a local file within the allowed base directory, creating parent directories as needed. Existing file contents are replaced.

Adds text to the end of a local file within the allowed base directory, creating the file and parent directories if needed.

Shows files and directories within the allowed base directory, with an optional glob pattern to filter results.

Shows which files in a Git working tree are staged, unstaged or untracked. Requires Git to be installed.

Shows unstaged changes in a Git working tree or changes already staged for commit. Requires Git to be installed.

Shows recent Git commits for reviewing a repository's history. Requires Git to be installed.

Stages selected files in a Git repository for a later commit. Requires Git to be installed.

Creates a commit from staged changes in a Git repository, using the supplied message. Requires Git to be installed.

Lists a Git repository's local branches. Requires Git to be installed.

Creates a Git branch from the current HEAD, with an option to check out the new branch. Requires Git to be installed.

Register a named Python function for batch processing. Map and filter functions take one item; reducers take an accumulator and an item and must be registered with is_reducer=True.

Apply a function to every item in a list, with a choice of error handling. Returns the results, success and error counts, and elapsed time.

Keep list items for which a function returns a truthy value, and report how many were kept or rejected.

Combine list items into one result using a function and a starting accumulator. Built-in reducers include sum, product, maximum, minimum and concatenation.

Split a list into items that pass a function's truth check and items that fail it, with a count for each group.

Divide a list into smaller batches of a specified size.

Combine corresponding items from multiple lists into records with specified keys.

Review success and error counts, total items and elapsed time from the most recent batch map run.

List the built-in functions and reducers available for batch processing.

Divide text into chunks by character count, estimated token count, sentences or paragraphs. Returns each chunk with its index, character count and token estimate.

Divide a list into batches of a specified size, returning each batch with its index and item count.

Split text at a custom separator, optionally merging the resulting parts up to a specified character limit.

Extract text windows using a chosen character length and advance distance. Returns each window's text and start and end positions.

Measure text length in characters, words, lines, sentences and paragraphs, with a rough token estimate based on about four characters per token.

Extracts a JSON value using a path such as user.name or items[0].price. Wildcard paths such as users[*].email collect matching values into a list.

Sets a value at a dot-notation path and returns the modified JSON. The replacement value must be supplied as a JSON-encoded string.

Lists a JSON object's top-level keys as a JSON array.

Filters a JSON array to the items whose specified key equals a supplied value.

Makes JSON easier to read by formatting it with adjustable indentation.

Combines two JSON objects, with values from the patch object replacing base values wherever their keys match.

Extracts a field from each item in a JSON array, optionally changing text case, converting values or applying another built-in transform.

Selects JSON array items by a field condition, including numeric comparisons, text matches and whether a value exists.

Summarises a JSON array with counts, totals, averages or other aggregates. It can also select the first or last field value, join values as text or return unique values.

Orders a JSON array by a field, ascending by default, with an option to reverse the order.

Collects JSON array items into groups that share a field value.

Combines nested lists into one JSON array, unpacking one level and keeping other items as they are.

Pairs items from two JSON arrays by position. Pairing stops at the end of the shorter array.

Keeps selected fields in each JSON object in an array, leaving non-object items unchanged.

Extracts part of a JSON array, including the start position and excluding the end position.

Checks for a regular expression match at the start of text, returning the matched text, positions and captured groups.

Finds the first regular expression match anywhere in text, with its positions and captured groups.

Collects non-overlapping regular expression matches. Patterns with capture groups return the captured text; this tool does not include match positions.

Collects regular expression matches with their start and end positions and captured groups.

Replaces regular expression matches in text, with support for references to captured groups. By default, it replaces every occurrence; a count limits replacements.

Divides text into a list of strings using a regular expression as the separator.

Extracts numbered and named capture groups from each regular expression match.

Checks whether a regular expression has valid syntax.

Escapes regular expression characters so supplied text can be matched literally.

Build a dictionary containing only selected keys, with a report of requested keys that were missing.

Build a dictionary with selected keys removed and report which keys were omitted.

Rename dictionary keys using an old-to-new name mapping, keeping keys that have no mapping.

Convert selected dictionary values to strings, integers, floats, booleans, lists, dictionaries or null, and report conversion errors.

Convert nested data into a flat dictionary with dot-separated keys, using numbered keys such as key.0 for array items.

Build a nested dictionary from flat, dot-separated keys.

Reshape a list of records by selecting or removing fields, renaming keys, converting value types and adding fields, in that order. Returns the transformed records and any errors.

Combine dictionaries recursively, with values from later dictionaries taking precedence when keys conflict.

Reports the current date and time in ISO 8601 format, with a timezone option.

Converts a date or time string to ISO 8601 format.

Formats a date and time string with a chosen strftime pattern.

Calculates the absolute difference between two dates and times, in seconds, minutes, hours or days.

Shifts a date and time forward or backward by a duration specified in days, hours, minutes or seconds.

Converts a date and time from its source timezone to a chosen destination timezone.

Converts a date and time string to a Unix timestamp measured in seconds since the epoch.

Converts a Unix timestamp to an ISO 8601 date and time string.

Turns a byte count into a readable size such as 1.2 MB, with binary units such as KiB and MiB available.

Turns a duration in seconds into readable time, such as 1h 1m 1s. A verbose option spells out the units.

Formats numbers with thousands separators and adjustable decimal places, such as 1,234,567.89.

Formats a value as a percentage. Values with an absolute value below 1 are multiplied by 100, so 0.87 becomes 87.0%.

Formats an amount with a currency symbol and numeric formatting, with support for currencies including USD, EUR, GBP and JPY.

Turns integers into English ordinal labels, such as 1st, 2nd, 11th and 21st.

Combines a count with the singular or plural form of a word, such as 3 items. Supports a custom plural form.

Shortens text to a specified maximum length, adding a suffix that defaults to ….

Pads text to a specified width with left, right or center alignment.

Turns a JSON array of objects into a plain-text table.

Draws a random sample from a list. Sampling allows repeated items when replacement is enabled; otherwise, the requested sample cannot exceed the list size.

Selects random items according to supplied weights, which are normalized automatically.

Samples a requested number of items from each group. Small groups are sampled with replacement, so items can repeat; empty groups remain empty.

Returns a randomly reordered copy of a list, leaving the original unchanged. A seed makes the order reproducible.

Shuffles a list and divides it into partitions using supplied proportions, returning the partitions and their sizes. The default is an 80/20 split.

Selects one random item from a list and returns its value and index.

Generates a requested number of random integers between lower and upper bounds, including both bounds.

Generates random floating-point values using lower and upper bounds and a chosen decimal precision. Rounding can produce the upper bound.

Summarize numeric data with its count, mean, median, standard deviation, variance, minimum, maximum, range and percentiles.

Group numeric values into histogram bins, showing each bin's range, count and frequency.

Measure the linear relationship between two numeric series with Pearson correlation, returning the coefficient, its square and an interpretation.

Rescale numeric values to the 0–1 range or standardize them as z-scores with mean 0 and standard deviation 1.

Identify unusually high or low values using the interquartile range or z-scores, with an adjustable detection threshold.

Smooth a numeric series with a simple or exponential moving average. The exponential method uses a smoothing factor of 0.3.

Count occurrences in categorical data and show each value's count and percentage.

Read a CSV or TSV file into JSON row objects, using the header row as field names.

Inspect the column names in a CSV or TSV file's header row.

Find CSV or TSV rows that meet a condition on one column, using equality, numeric comparisons, text containment or a text prefix. Matching rows are returned as JSON.

Calculate a count, sum, average, minimum, maximum or unique values for a column in a CSV or TSV file.

Write JSON row objects to a CSV file and return the number of rows written. Writing nonempty rows replaces any existing contents at the destination.

Fill a text template's ${variable} placeholders with supplied values. Missing variables cause an error by default; optional non-strict rendering leaves them in place.

Store a template by name for reuse during the session. Saving under an existing name replaces that template.

Fill a previously saved template with supplied values. The template must already exist; missing variables cause an error by default.

Find the variable names used in a template and return a sorted list without duplicates.

Check whether supplied variables cover a template's requirements, identifying missing variables and extra values.

Browse saved templates alphabetically, with the variables each one uses.

Retrieve a saved template's original text and variable names.

Delete a saved template by name.

Read a cached value by key. Returns null when the key is missing or its value has expired.

Save a value under a cache key, persisted to disk. An expiry time controls how long it remains valid; ttl_seconds=null disables expiry.

Remove one entry from the cache by key.

Delete every entry in the cache.

Check active and expired cache entry counts, plus cache hits and misses for the session.

Retrieves rows from the configured local SQLite database with a SQL SELECT query and displays them as a table.

Runs SQL that changes the local SQLite database, including inserting, updating or deleting rows and creating or dropping tables.

Shows the user-created tables in the configured local SQLite database.

Retrieves a SQLite table’s CREATE TABLE statement to inspect its structure.

Saves a fact under a key in persistent memory for later conversations. Saving to an existing key replaces its value.

Searches saved facts by keywords or phrases to retrieve information from persistent memory.

Deletes the fact stored under a specified key from persistent memory.

Creates a named document index in memory, optionally restricting which document fields are indexed.

Adds JSON documents to a named index, creating the index automatically if it does not exist.

Finds documents matching a keyword query and ranks them by relevance, returning a score with each result. Search can be restricted to one field.

Removes a document from an index using its _id.

Browses documents in a named index in pages.

Shows a named index's document count, token count and indexed fields.

Deletes a named index and the documents it contains.

Lists document indexes with their document counts.

Store an embedding and optional metadata in a named, in-memory vector store. Supply an ID or let the tool generate one; this stores an existing embedding rather than generating it.

Find vectors nearest to a query embedding in a named store, using cosine similarity, Euclidean distance or dot product. Results include IDs, scores and metadata, with optional limits on result count and minimum score.

Retrieve a vector and its metadata from a named store using its ID.

Remove a vector from a named store by ID and report whether an entry was deleted.

Browse the IDs in a named vector store in pages.

Inspect a named vector store’s vector count, dimensions and configured capacity.

Delete a named vector store and every vector it contains.

List named vector stores with their vector counts and dimensions.

Run Python or Bash code in a subprocess and return its output and errors. Network access is not blocked at the operating-system level, even when allow_network is false.

Set a key and value in a named configuration store, including keys written with dot notation.

Read a value from a named configuration store, optionally converting it to an integer, float, Boolean, string or JSON value.

Add multiple default values to a named configuration store, setting only keys that are not already present.

Load environment variables into a named configuration store, optionally selecting variables by prefix.

List the keys in a named configuration store, optionally filtered by prefix.

Delete a specific key from a named configuration store.

Export every configuration value in a named store as a JSON object.

Check that required configuration keys exist and report any missing keys.

Delete an entire named configuration store.

Shows the named configuration stores and how many keys each contains.

Reads one environment variable. Names identified as sensitive, such as those containing KEY, TOKEN or SECRET, return [hidden] instead of their values.

Sets an environment variable in the current process and makes it available to subsequently started child processes. The parent shell is unaffected.

Shows environment variables as a JSON object, optionally filtered by a name prefix. Variables identified as sensitive are excluded.

Checks whether required environment variables are set and lists those present or missing.

Loads key=value pairs from a .env file into the current process environment without replacing existing variables. Reports loaded and skipped names, plus errors.

Executes a command on the server's host and returns its output, errors and exit code. Commands can change files or system state; pipes and redirection require shell mode.

Executes a multiline script on the server's host through a temporary file and returns its output, errors and exit code. The interpreter must be installed; scripts can change files or system state.

Locates an executable on the server's PATH and returns its full path, or reports that it was not found.

Registers a subscriber name for a topic in the in-process event bus. The topic * matches events across all topics.

Removes a subscriber’s registration from a specified event-bus topic.

Records an event in the in-process bus, returns its ID and matching topic and wildcard subscriber names, and increments their event counts.

Retrieves a topic’s most recent retained events, ordered from oldest to newest within the selection.

Shows event-bus topics with their registered subscribers and retained event counts.

Shows the subscriber names registered directly to a specified topic.

Shows total published events and the event counts recorded for each subscriber.

Deletes a topic’s event history and subscriber registrations. Omitting the topic clears all topics and also resets the bus’s event and subscriber counts.

Creates a named directed graph for modeling dependencies or relationships.

Adds a node to a named graph, with optional metadata.

Adds a directed connection from one node to another, creating either endpoint if it does not yet exist.

Removes a directed connection between two nodes in a graph.

Removes a node from a graph along with every incoming and outgoing connection.

Lists the names of the nodes in a graph.

Lists a graph's directed connections, identifying the source and destination of each.

Checks whether a directed graph contains a cycle.

Lists graph nodes in topological order, with each connection's source before its destination. Returns an error if the graph contains a cycle.

Finds the shortest directed path between two nodes, reporting the route and its length or whether the destination is unreachable.

Finds all nodes from which a specified node can be reached by following directed connections.

Finds all nodes reachable from a specified node by following directed connections.

Shows a graph's node count, connection count and node names.

Deletes a named graph, including its nodes and connections.

Lists the available graphs with the number of nodes and connections in each.

Acquires a named lock for an owner to coordinate concurrent operations. Supports automatic expiry and a blocking wait, and reports whether the lock was acquired.

Releases a named lock held by the specified owner. Only the owner holding the lock may release it.

Attempts to acquire a named lock without waiting. Immediately reports whether it succeeded and, if the lock is held, who holds it.

Shows whether a named lock is held, with its owner and remaining lifetime when available.

Lists the locks currently held and reports their count.

Releases every lock held by a specified owner and reports which locks were released.

Forces a named lock to release, even when another owner holds it.

Reports lock acquisitions, contention and the number of locks currently held.

Creates a named queue that serves items in arrival order, reverse arrival order or priority order. Lower priority numbers come first; a maximum size of 0 leaves the queue unlimited.

Adds a JSON value to a named queue. In a priority queue, lower priority numbers are served first.

Takes the next item out of a queue and reports the remaining size, or indicates that the queue is empty.

Shows the next item in a queue while leaving it in place.

Reports how many items are waiting in a named queue.

Empties a named queue while keeping the queue itself.

Removes a named queue entirely.

Lists active queues with their types and current item counts.

Saves a task prompt on a recurring interval or for a single ISO date and time, replacing any task with the same ID. It records the schedule; processing due tasks requires a separate scheduler check.

Lists saved tasks and their next scheduled times.

Removes a task from the saved schedule by its ID.

Returns tasks that are due for another process to execute. It also advances recurring tasks to their next scheduled time and removes one-time tasks from the schedule.

Deletes every task from the saved schedule.

Create a named state machine with a starting state and valid states. Allowed transitions are added separately.

Define a permitted move from one state to another in an existing state machine.

Move a state machine to a requested state if the transition is allowed, or return an error if it is not.

See a state machine's current state and the states it can move to next.

Check whether a state machine can move to a particular state without making the transition.

Review recent state transitions, including their sequence numbers, starting and ending states, and timestamps.

Reset a state machine to a specified valid state and clear its transition history.

Inspect a state machine's defined states, current state, permitted next states and recorded transition count.

Delete a named state machine.

Browse existing state machines with their current states and recorded transition counts.

Register a named sequence of processing steps, with optional retries, conditions and error handling.

Execute a registered workflow on input data and return its output, step results, elapsed time and success status.

Register Python source as a custom workflow step function. The source must define def step(value, ctx): ....

List registered workflows.

Inspect a workflow’s full definition by name.

Remove a registered workflow by name.

Review workflow execution history, including total, successful and failed runs and recent executions.

List the built-in processing functions available for workflow steps.

Sends HTTP requests to APIs, webhooks and other endpoints, returning the response status, headers and body. Supports custom authentication headers and request bodies; POST, PUT, PATCH and DELETE requests can change remote data.

Creates a named rate limiter using a fixed window, sliding window or token bucket. Sliding windows avoid bursts at window boundaries; token buckets allow a configurable burst while replenishing capacity over time.

Checks whether a request fits within a named rate limit without using its allowance, and reports remaining capacity and reset or wait timing.

Records a request against a named rate limit, using its allowance if capacity remains. Reports whether consumption succeeded.

Checks a named rate limit and consumes allowance when the request is allowed, combining both actions in one call. A denied request consumes no allowance.

Shows a named rate limiter's usage, remaining capacity and algorithm settings.

Restores a named rate limiter's counters or tokens to their initial state.

Removes a named rate limiter and its tracked state.

Lists active rate limiters with their names, algorithms and current state.

Sends an HTTP request and retries network failures, rate-limit responses and server errors, with increasing delays and optional random variation. Write requests may repeat, so use operations that are safe to retry.

Runs a shell command on the server's host and retries after a non-zero exit, with increasing delays. The command may run more than once, including any file or system changes it makes.

Shows retry activity totals: calls, retries, successes and failures.

Creates a named circuit breaker, or resets an existing one. It blocks requests after a configured number of consecutive failures, then permits a trial request after a waiting period.

Sends an HTTP request through a named circuit breaker. While the breaker is open, it rejects the call without sending a request; after the waiting period, it permits a trial call.

Shows a named circuit breaker's state, failure count, failure threshold and recovery waiting period.

Clears a named circuit breaker's failure count and closes it, allowing requests to pass through again.

Start a temporary HTTP listener for one POST webhook and return its body and headers. The listener stops after receiving a request or timing out; the sender must be able to reach its listening address and port.

Create a named alert rule for numeric comparisons, ranges, text matches, empty values or truth checks, with an info, warning, error or critical severity.

Check a value against a named alert rule and report whether it fired, together with its severity and timestamp.

List the defined alert rules and their count.

Look up the full definition of a named alert rule.

Delete a named alert rule.

Review fired alerts, with filters for rule name and severity and a limit on the number returned.

Delete the alert history for one rule, or the entire alert history when no rule is specified.

See how often each alert rule has fired and when it last fired, plus the total number of firings.

Record a structured audit event with its type, actor, resource, severity and outcome.

Find audit events by event type, actor, resource, severity, outcome, time range or text.

Retrieve a single audit event by its ID.

Summarize audit events by type, actor, resource, severity and outcome.

Export audit events as JSON Lines, optionally filtered by event type and time range.

Delete audit events, optionally restricting deletion to events older than a specified timestamp. Reports how many were deleted and how many remain.

List the distinct event types recorded in the audit log.

Group audit events by hour or day to see activity over time.

Adds a value to a named counter, creating it if needed. The default increment is 1; results include the total, number of increments and smallest, largest and latest increments.

Records a metric's current value, such as queue depth or active connections, replacing the previously recorded value.

Starts measuring elapsed time and returns an ID for stopping that timer. Separate timers can run under the same name.

Stops a timer by its ID and records the duration in milliseconds, updating the named timer's count, total, minimum, maximum and average durations.

Retrieves the recorded state of a named metric, or reports an error if it does not exist.

Lists recorded metrics by name, with each metric's type, in alphabetical order.

Returns the recorded state of every metric, organised by metric name.

Clears a named metric's accumulated values. Omitting the name resets every metric; the result reports how many were reset.

Exports recorded metrics as JSON or Prometheus text, using JSON by default.

Announces a task result or event through a desktop notification when available and records it in the notification log. If desktop delivery is unavailable, it logs the message only.

Displays a desktop notification through notify-send on Linux or osascript on macOS. Reports when desktop delivery is unavailable.

Appends a notification to a JSONL log file without requiring a desktop display.

Sends a terminal bell signal to draw attention to an event.

Retrieves the most recent entries from a notification log as a JSON array.

Start a named stopwatch or restart an existing one.

Stop a named stopwatch and retrieve its elapsed time in milliseconds and seconds, along with recorded lap splits.

Check a stopwatch's elapsed time and whether it is running, without stopping it.

Record a stopwatch lap split, with its duration, lap number and total elapsed time.

Reset a stopwatch and immediately start it again.

Remove a named stopwatch.

Browse stopwatches with their elapsed times and recorded lap splits.

Start a named countdown for a specified number of seconds.

Check how much time remains on a countdown and whether it has expired.

Browse countdowns with their remaining times.

Measure a shell command over repeated runs and report average, minimum, maximum and total execution time. The command is executed on every run.

Convert text to speech. The system speech engine plays it aloud; a configured HTTP speech server receives the text and returns audio that is saved as an MP3 file.

Creates a cryptographically secure random token in hexadecimal. The length is measured in random bytes: the default 32 bytes produces 64 hexadecimal characters.

Calculates a string’s hash as a hexadecimal digest, using SHA-256 by default. Also supports MD5, SHA-1, SHA-224, SHA-384 and SHA-512.

Creates a hexadecimal HMAC signature from data and a shared secret, for uses such as signing webhook payloads.

Checks an HMAC signature against the original data and shared secret, returning whether it matches. Uses constant-time comparison.

Creates a random version 4 UUID for use as an identifier.

Converts text to Base64, with an option for URL-safe encoding.

Converts Base64-encoded text back to its original string, including URL-safe Base64 when selected.

Creates random bytes for values such as nonces, salts or initialization vectors, returning them in hexadecimal.

Check an email address against simplified format rules. This checks syntax only; it does not confirm that the mailbox exists or can receive mail.

Check that a URL has a host and an allowed scheme. HTTP and HTTPS are allowed by default; the check does not establish whether the destination is reachable.

Check an IPv4 or IPv6 address and identify its version, whether it is private and whether it is a loopback address.

Check a UUID string and identify its version and variant. Accepts UUIDs with or without hyphens.

Parse a JSON string to check its validity and return the parsed value. Can also check for required top-level keys when the value is an object.

Check whether a number falls within specified minimum and maximum bounds, including the boundary values.

Search a string for a regular expression match and return captured groups. Use anchors in the pattern when the whole string must match.

Check whether a string or list meets minimum and maximum length limits, including the boundary lengths.

Check a value against an expected type: string, integer, float, number, boolean, list, dictionary or null. Reports the actual and expected types.