get_contextLABSCO SUMMARY
The orientation an agent would otherwise spend ten file reads building.
One call returns the architecture facts and the constraints other engineers wrote down — the two things a coding agent cannot infer from the file in front of it. The attestation at the end is what makes its work reviewable: which rules it checked, which elements moved, and where it could not tell.
What it is
A context source for coding agents. It answers what this project is, which rules cover the file about to be edited, what changed recently, what is currently flagged, and what the working diff actually touches.
What you get
- A project digest — architecture, tech stack, data model, rules — retrievable whole or drilled into by path
- Per-file context: the architecture elements a file belongs to and the constraints that apply to it
- The current findings, with a short curated shortlist of what needs attention first
- A change log of which architecture elements and rules moved, when, and by whom
- The blast radius of the working diff: touched elements, triggered rules, files that map to nothing
- A two-phase commit attestation: draft it, then write it with your own status per rule
- Cross-session memory, so what an earlier session learned is readable by the next one
- A hosted server by URL, or a local one launched through the CLI
Requirements
A gjalla project. The hosted server is a URL plus a browser approval; the local one runs through the CLI, which already holds the configuration, so the MCP config carries no keys.
Setup effort
Follow the published steps — pip install gjalla[mcp]
USE CASES
01
Starting a session on an unfamiliar codebase
Pull the project digest and the current findings before touching anything, instead of grepping your way to a mental model.
02
Editing a file without breaking a rule nobody mentioned
Ask what the file belongs to and which constraints apply to it before the first edit.
03
Working out what a change actually touches
Read the diff's blast radius — architecture elements, triggered rules, unmapped files — before you commit rather than after CI does.
04
Leaving a record a reviewer can trust
Draft the attestation, then write it with your own status per rule, marking the ones you could not determine instead of claiming compliance.
05
Picking work back up after a break
List what moved since a date and what is currently flagged, rather than reading the whole history.
21 tools
Listed in project docs. Your client may show a different list.
get_file_contextget_statusget_impactprepare_attestationget_changesget_findingsget_element_detailsget_source_of_truth_docget_project_skillsget_memoriesdiscover_systemsget_system_contextget_system_architectureget_system_mappingsdiscover_projectssetupsyncadd_ruleadd_skillsave_memoryBEFORE YOU INSTALL
What to know first
Tools you get
21 tools named in the project's documentation. save_memory, the write side of get_memories, has no reference entry.
Where it runs
Of the 6 documented clients, 3 run the server on your machine and 3 connect to a hosted endpoint.
INSTALLATION
Connect your client
Server setup
Setup
TERMINAL
pip install gjalla[mcp]
gjalla setup mcpHosted endpoint
ENDPOINT URL
https://gjalla.io/mcp/streamAdd this hosted endpoint to your MCP client, then sign in.
MCP Clients
Claude Code, Cursor
JSON
{
"mcpServers": {
"gjalla": {
"command": "gjalla",
"args": [
"mcp",
"serve"
]
}
}
}Claude.ai
Settings
Integrations (or Connectors, depending on plan)
add a custom MCP server with the URL
approve the OAuth request.
CLAUDE.AI · URL
https://gjalla.io/mcp/streamLovable
Settings → Integrations → Custom MCP → paste the URL → Connect
approve in the gjalla popup.
LOVABLE · URL
https://gjalla.io/mcp/streamChatGPT
Open custom-connectors / MCP settings (visibility depends on plan)
add a connector with the URL
approve the OAuth request.
CHATGPT · URL
https://gjalla.io/mcp/streamAny MCP client — uvx
ANY MCP CLIENT — UVX · JSON
{
"mcpServers": {
"gjalla": {
"command": "uvx",
"args": [
"--with",
"gjalla[mcp]",
"gjalla",
"mcp",
"serve"
]
}
}
}
