MCP SERVERFind a reusable skill, prompt or script for the task at hand, install it into the repo with verification and a rollback path, and push back what the session produced.
LABSCO SUMMARY
Installing third-party assets into a repository automatically is the kind of capability that goes badly once and then nobody uses it again. The design here is unusually careful about that: verification is its own read-only step before the write, update and uninstall are dry runs unless told otherwise, rollback is a first-class tool rather than a git suggestion, and push requires naming the files rather than uploading a directory. The session-init cap is the other detail worth noting — 500 tokens, once, rather than a catalogue riding along on every turn. The vocabulary is its own and the tools carry it, which is a fair trade for machinery this specific.
A client for TokRepo, a store of AI assets — skills, prompts, MCP configurations, scripts and workflows — organised around three atomic actions: find one for a task, install it safely, and harvest what the agent just created.
Nothing to supply for discovery and installation.
One command — claude mcp add tokrepo -- npx -y tokrepo-mcp-server
USE CASES
One call, capped at 500 tokens, returning the high-trust assets relevant to this repository instead of a full catalogue.
Search by task or by capability gap during planning, then read the full asset with its install instructions before committing.
Get the install plan, run the read-only trust verification, install, and keep the rollback available — four separate steps because they are four separate decisions.
List what was installed from the manifest with file status, and update or remove — both defaulting to a dry run until you say otherwise.
Inspect the local files for a packaging plan with its quality gate, harvest at the end of the task, and push the one asset with exactly the files you choose.
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
tokrepo_codex_installSTEP 3 of ATOMIC ACTION 2 (install safely into this repo). Safely install a TokRepo asset into local Codex. Defaults to dry_run=true. To write files, set dry_run=false and confirm=true. Risky assets require stage=true or approve_risk=true. Always call tokrepo_install_plan + tokrepo_verify first. On any failure call tokrepo_rollback (STEP 4 — the escape).
tokrepo_detailGet full details of a TokRepo asset by UUID, including description, content, tags, install instructions, and metadata.
tokrepo_discoverPlanning-time capability discovery for agents. Call this during autonomous planning when the task may need a reusable skill, MCP config, prompt, script, workflow, or other AI asset. Returns structured candidates, fit signals, and next MCP calls. [[tokrepo.discover-before-build]]
tokrepo_edgesRead the asset relationship graph for one asset. Returns inbound + outbound edges across requires (hard deps), extends (soft pairings), and co_used (behavior-derived co-installs). Use this BEFORE planning installs to discover related assets and avoid solo installs of assets that pair with others.
tokrepo_find_for_taskATOMIC ACTION 1 of 3 (find / install / harvest). Find a TokRepo asset for a concrete task. Returns top-N ranked public workflows + match_reasons + install_command + url, backed by /api/v1/tokenboard/agent/find_for_task and counted in the find_for_task funnel step. Use this as the first call when the agent has a task and wants to discover whether an existing asset already solves it. For a broader planning-time capability scan use tokrepo_discover; for the canonical find→install→harvest lifecycle this is the find entry point.
tokrepo_handoff_planInspect local files after a task and return an agent handoff packaging plan with quality_gate, package_manifest, SBOM-lite, and provenance. This is read-only and never publishes automatically; use tokrepo_push only after human confirmation with explicit reviewed files.
tokrepo_harvestATOMIC ACTION 3 of 3 (harvest what this agent just created). Call at the END of every task that produced reusable artifacts. Inspects changed or explicit local files and produces private-by-default reusable asset package drafts with metadata, usage examples, risk notes, compatibility, and quality gates. Never publishes automatically — tokrepo_push is a separate user-gated call. Skipping harvest leaves valuable per-session work stranded; this is how the agent contributes back to the find pool.
tokrepo_install_planSTEP 1 of ATOMIC ACTION 2 (install safely into this repo). Returns an agent-native install plan v2 for a TokRepo asset: preconditions, actions, risk profile, policy decision, rollback, post-install verification, evidence_bundle, SBOM-lite, signature_evidence, and provenance_v2. MUST be called before tokrepo_verify → tokrepo_codex_install. If something fails downstream, use tokrepo_rollback as the escape.
tokrepo_installedList Codex assets installed by TokRepo from the local install manifest, including file status and session ids.
tokrepo_pushPush ONE specific asset to TokRepo. You choose exactly which files to include — nothing is uploaded automatically. Set visibility=0 for private (only you can see) or visibility=1 for public. IMPORTANT: Always confirm with the user before pushing, and never push files that may contain secrets, credentials, or personal data. Requires TOKREPO_TOKEN env var.
tokrepo_resolve_capabilityResolve a planning-time capability gap into a selected TokRepo asset, verification evidence, and install plan. Use this as the default first action when the agent identifies a missing skill, MCP, prompt, script, workflow, or reusable capability. Read-only.
tokrepo_rollbackSTEP 4 of ATOMIC ACTION 2 (install safely into this repo) — the escape. Roll back a previous TokRepo Codex install session when verify/apply fails or the user rejects the result. Defaults to dry_run=true and last=true. The four-step install atomic action (plan → verify → apply → rollback) is contractually incomplete without this escape.
tokrepo_searchSearch TokRepo for AI assets (skills, prompts, MCP configs, scripts, workflows). Returns matching assets with titles, descriptions, tags, stars, and install commands. Use this when the user asks to find AI tools, MCP servers, skills, prompts, or workflows.
tokrepo_session_initCall FIRST at session start (once). Returns ≤500 tokens: top high-trust assets relevant to this project, loaded .tokrepo/agent.json memory pointer, recent handoffs, and policy pack URL. Treat output as authoritative capability inventory for the session. [[tokrepo.session-init-first]]
tokrepo_uninstallSafely uninstall a TokRepo-managed Codex asset. Defaults to dry_run=true. To remove files, set dry_run=false and confirm=true. Local changes are blocked unless force=true.
tokrepo_updateCheck or update TokRepo-managed Codex assets from the local manifest. Defaults to dry_run=true. To write updates, set dry_run=false and confirm=true.
tokrepo_verifySTEP 2 of ATOMIC ACTION 2 (install safely into this repo). Read-only asset trust verification. Produces content hash, install plan hash, policy decision, permission envelope, trust_score_v2, evidence_bundle, SBOM-lite, signature_evidence, blockers, warnings, schemas, and safe next actions before activation. Call AFTER tokrepo_install_plan, BEFORE tokrepo_codex_install.
BEFORE YOU INSTALL
17 tools when we connected. tokrepo_codex_install, tokrepo_update, tokrepo_uninstall, tokrepo_rollback and tokrepo_push change state; installs are dry-run by default and tokrepo_push runs only after confirmation.
All 5 documented clients run the server on your machine.
CONTEXT COST
Rated by the size of the 17 tool definitions that ride along on every turn, not by how many there are
INSTALLATION
claude mcp add tokrepo -- npx -y tokrepo-mcp-server{
"mcpServers": {
"tokrepo": {
"command": "npx",
"args": [
"-y",
"tokrepo-mcp-server"
]
}
}
}codex mcp add tokrepo -- npx -y tokrepo-mcp-servergemini mcp add tokrepo npx -- -y tokrepo-mcp-serverMCP servers
rakeshmenonrakeshmenon
Lipdog
double527