Skip to main content
Glama

Server Configuration

Describes the environment variables required to run the server.

NameRequiredDescriptionDefault
KEYSTONE_CONFIGNoPath to the configuration YAML file (default: .keystone/context.yaml).keystone/context.yaml

Instructions

Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.

This server publishes no instructions, or was last inspected before Glama recorded them.

Capabilities

Features and capabilities supported by this server

Protocol revision2025-11-25

CapabilityDetails
tools
{
  "listChanged": true
}
logging
{}
prompts
{
  "listChanged": false
}
resources
{
  "subscribe": false,
  "listChanged": false
}
extensions
{
  "io.modelcontextprotocol/ui": {}
}
experimental
{}

Tools

Functions exposed to the LLM to take actions

NameDescription
keystone_get_contextC

Full envelope (rules + reasoning + skills + commands) for a topic.

keystone_list_topicsA

List configured topics. Pass tag to filter.

keystone_harness_bootstrapA

Create .keystone/harness/ and (by default) materialize the shipped template tree.

Idempotent — existing subdirs and files are reported in skipped, never overwritten. Call this once per project before scaffolding individual guides / sensors / actions / playbooks.

Pass materialize_templates=False to get the bare-bones directory layout only (no shipped state ledgers, sensors, actions, or playbooks). The full tree is the recommended default; opt out only for advanced use cases.

keystone_new_guideA

Scaffold a new guide markdown file under .keystone/harness/guides/.

tier ∈ iron-law | golden | rules. Strictness cascade: iron-law (can never be violated) > golden (hard rule; deviation requires explicit reasoning) > rules (regular rule; golden rules can override).

keystone_new_sensorA

Scaffold a new sensor + its matching implementation.

Sensors are blocking rules. mode selects how the agent runs them:

  • computational (default) → stamps scripts/<name>.sh (shell). Agent runs via Bash; exit 0 = pass, non-zero = fail.

  • inferential → stamps prompts/<name>.md (markdown). Agent reads the prompt and performs the reasoning task it describes (e.g. code review, security review). Reports PASS / FAIL.

kind ∈ lint | type | test | build | drift | coverage | computational | domain | custom — informational category.

keystone_new_scriptA

Scaffold a shell script under .keystone/harness/scripts/<name>.sh.

Use this to drop a script body without a sensor wrapper, or to refresh an existing script (with force=True). New scripts are chmod +x. Most projects scaffold sensors via keystone_new_sensor which stamps the matching script automatically.

keystone_new_promptA

Scaffold a prompt markdown under .keystone/harness/prompts/<name>.md.

Used by inferential sensors — the agent reads the prompt and performs the reasoning task it describes. Most projects scaffold inferential sensors via keystone_new_sensor(mode="inferential") which stamps the matching prompt automatically.

keystone_new_skillA

Scaffold .keystone/harness/skills/<name>/SKILL.md.

Skills are the FastMCP-native primitive for agent-discoverable procedural how-to. Each subdirectory containing a SKILL.md becomes a discoverable skill, surfaced as skill://<name>/SKILL.md and auto-loaded by agent runtimes (Claude Code, Cursor, etc.).

Manager-authored skills are named keystone-<slug>; the scaffolder prepends keystone- if missing.

keystone_new_actionA

Scaffold a new action markdown under .keystone/harness/actions/<name>.md.

Actions are short, focused operations the agent walks during a task — spec, orient, implement, verify, review, learn, audit, release. They complement playbooks (which orchestrate actions into a flow) and skills (which expose procedural how-to via the FastMCP skill:// scheme).

keystone_new_playbookB

Scaffold a new playbook markdown under .keystone/harness/playbooks/<name>.md.

Playbooks orchestrate multiple actions into a higher-level flow with explicit phase gates: task, bootstrap, audit, install, verify, doctor, patch, release.

keystone_new_corpusA

Scaffold a new corpus markdown under .keystone/harness/corpus/<name>.md.

Corpus entries are reasoning / background context — domain notes, architecture decisions, idioms. Not constraints (those go in guides/) and not procedures (those go in actions/, playbooks/, skills/).

keystone_new_adapterB

Scaffold a per-agent adapter directory under .keystone/harness/adapters/<agent>/.

keystone_target_addA

Install the agent's menu file(s) at the project root.

Menu files (CLAUDE.md, AGENTS.md, etc.) point the agent at .keystone/harness/ and at this MCP server. They are thin pointers, not content — the single source of truth lives in the harness. Phase 19 overlay semantics: only the region between <!-- BEGIN KEYSTONE --> and <!-- END KEYSTONE --> is rewritten; pre-existing user content is preserved.

keystone_apply_patchesA

Apply every pending shipped patch to the project harness.

Patches are forward-only. Files modified by the user since the previous shipped version are skipped and reported as conflicts; the user resolves them by hand. Today no patches ship — the call reports an empty applied list. Future releases populate templates/patches/<version>/.

Prompts

Interactive templates invoked by user choice

NameDescription
keystone_bootstrapSeed the bootstrap workflow: analyze the codebase and fill the project's state ledgers under `.keystone/harness/corpus/state/`.
keystone_taskSeed the task workflow on a unit of work. Walks spec → orient → implement → check-drift → verify → review. Pause for explicit user acceptance between phases.
keystone_auditSeed the dual-flywheel audit: learning + pruning.
keystone_learnSeed the learn workflow: capture a finding into learning/inbox/.

Resources

Contextual data attached and managed by the client

NameDescription
context_list_resource
harness_status_resource
harness_options_resource
harness_verify_resourceCascade-engine resolution report for the current harness (Phase 20). Surfaces unreachable items, canonical violations, required gaps, and non-canonical conflicts. Read-only — no side effects.
harness_doctor_resourceFull audit report — cascade verify + path conformance + ambient-load budget proxy. Read-only.
harness_patch_pending_resourcePending shipped patches against the current harness (Phase 21). Lists files that would be written if the consumer ran the patch playbook, plus files skipped because the consumer has modified them since the previous shipped version.
harness_budget_resourceAmbient-load budget report for the project harness (Phase 27). Per-port totals, top hot files, approximate-token counts via a deterministic word-count proxy, and (when the cascade marks items unreachable) a `cascade_excluded` block.

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/tacoda/keystone-mcp'

If you have feedback or need assistance with the MCP directory API, please join our Discord server