io.github.kultar/librarian
by kultar
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| LIBRARIAN_HOST | No | Address the listener for agents (A2A and MCP) binds to. Default: 127.0.0.1. | |
| LIBRARIAN_PORT | No | Port for the agent listener (A2A and MCP). Default: 8080. | |
| LIBRARIAN_MODEL | No | The model for plain-language requests (fetch) and for learning an API that has no spec, e.g. anthropic:claude-sonnet-5-5. Unset: those two are unavailable. | |
| LIBRARIAN_A2A_URL | No | The A2A address advertised in the agent card when agents and people reach librarian at different addresses. Default: public_url. | |
| LIBRARIAN_CALLERS | No | JSON list of caller objects with agent, key_ref, apis, can_onboard, write_apis, owner, approval_wait_seconds and limits. With no callers configured the server is open. | |
| LIBRARIAN_REPLICAS | No | How many replicas run. Above 1, librarian refuses to start without Postgres and the AWS secret store. Default: 1. | |
| LIBRARIAN_MODEL_API | No | The format spoken to the gateway: openai or anthropic. Default: openai. | |
| LIBRARIAN_AUDIT_FILE | No | The audit log, one JSON object per line. Unset: standard output. | |
| LIBRARIAN_PAGES_HOST | No | Address of the separate listener for the pages and console. Default: 127.0.0.1. | |
| LIBRARIAN_PAGES_PORT | No | Serve the pages and console on their own listener. Needed with approvers.mode: trusted_header. | |
| LIBRARIAN_PUBLIC_URL | No | Where approvers' browsers reach the pages; every link librarian sends starts with it. Default: http://127.0.0.1:8080. | |
| LIBRARIAN_MCP_ENABLED | No | Serve the skills as MCP tools at /mcp. Default: true. | |
| LIBRARIAN_SECRETS_DIR | No | With file: the folder holding one 0600 file per secret. Default: .librarian/secrets. | |
| LIBRARIAN_DATABASE_URL | No | Where everything is kept. Default: sqlite+aiosqlite:///.librarian/librarian.db. Use postgresql+asyncpg://... for several replicas; null keeps it in memory. | |
| LIBRARIAN_OTEL_ENABLED | No | Send traces and metrics with OpenTelemetry; endpoint comes from the standard OTEL_EXPORTER_OTLP_* variables. Default: false. | |
| LIBRARIAN_REGISTRY_DIR | No | Registry entries written by hand as YAML. With a database they are imported into it. Default: registry. | |
| LIBRARIAN_MODEL_API_KEY | No | The key for the model gateway. It is the only key ever sent there. | |
| LIBRARIAN_PUBLISH__REPO | No | The record repository, owner/name. | |
| LIBRARIAN_PUBLISH__TYPE | No | none or github; github writes every approved entry and decision to a repository as a commit. Default: none. | |
| LIBRARIAN_FALLBACK_MODEL | No | Tried once when the model fails or declines a request. | |
| LIBRARIAN_INTERNAL_HOSTS | No | JSON list of host name patterns agents may learn APIs on although they resolve to private addresses, e.g. ["*.internal.example.com"]. | |
| LIBRARIAN_MODEL_BASE_URL | No | An OpenAI-compatible gateway such as LiteLLM, instead of the provider. | |
| LIBRARIAN_NOTIFIER__TYPE | No | Where requests for a person go: log, slack or webhook. Default: log. | |
| LIBRARIAN_PLANNER_OUTPUT | No | How the planner gets structured answers: tool, native, prompted, or auto. Default: auto. | |
| LIBRARIAN_PUBLISH__TOKEN | No | A token for the record repository with contents and pull requests write permission. Prefer publish.github_app. | |
| LIBRARIAN_WRITES__GRANTS | No | JSON list of config grants for unattended agents, with agent, api, operation, principal, pinned_args, max_uses_per_day and notify_owner. | |
| LIBRARIAN_APPROVERS__MODE | No | How the pages know who is deciding: token (whoever holds the link) or trusted_header (a sign-in proxy in front). Default: token. | |
| LIBRARIAN_LEARNING__MODEL | No | A different model for learning than for fetch. | |
| LIBRARIAN_RECIPES_ENABLED | No | Save a successful plain-language request as a recipe and replay it without the model. Default: true. | |
| LIBRARIAN_SECRETS_BACKEND | No | Secret backend: file or aws. Default: file. | |
| LIBRARIAN_SIGNING_KEY_REF | No | The secret holding the key that signs links for people. Default: librarian/signing-key. | |
| LIBRARIAN_APPROVERS__GROUP | No | The group whose members may approve. | |
| LIBRARIAN_PRINCIPALS__MODE | No | Whether the person an agent names must be proven: off, optional or required. Default: off. | |
| LIBRARIAN_PUBLISH__API_URL | No | The GitHub API address, for GitHub Enterprise Server. Default: https://api.github.com. | |
| LIBRARIAN_SUBSTRATE__RULES | No | JSON list of substrate rule objects with match fields, agent, apis, can_onboard, write_apis, owner, approval_wait_seconds and limits. | |
| LIBRARIAN_ALLOW_PROVISIONAL | No | Whether a learned API can be read before a person approves it. Default: true. | |
| LIBRARIAN_MCP_ALLOWED_HOSTS | No | Host headers accepted on /mcp. Empty means any; every MCP request still needs a caller's key. | |
| LIBRARIAN_OTEL_SERVICE_NAME | No | The service name carried by telemetry. Default: librarian. | |
| LIBRARIAN_SUBSTRATE__ISSUER | No | Issuer of actor tokens, e.g. https://idp.ate-system.svc. Setting it turns Agent Substrate/kagent authentication on. | |
| LIBRARIAN_WRITES__APPROVERS | No | Who may approve a write: requester, group, or requester_or_group. Default: requester_or_group. | |
| OTEL_EXPORTER_OTLP_ENDPOINT | No | Standard OpenTelemetry OTLP endpoint used when otel_enabled is true. | |
| LIBRARIAN_APPROVERS__CONSOLE | No | The console: signed_in, open (no sign-in: a laptop) or off. Default: signed_in. | |
| LIBRARIAN_MAX_RESPONSE_BYTES | No | An API answer longer than this is cut off and marked truncated. Default: 2000000. | |
| LIBRARIAN_ONBOARDING_ENABLED | No | Whether agents may teach librarian new APIs at all. Default: true. | |
| LIBRARIAN_PRINCIPALS__HEADER | No | The header carrying the person's token. Default: X-Principal-Token. | |
| LIBRARIAN_PUBLISH__DIRECTORY | No | The folder in the record repository. Default: registry. | |
| LIBRARIAN_SUBSTRATE__CA_FILE | No | CA bundle for reaching the issuer, such as the trust bundle of its signer in a cluster. | |
| LIBRARIAN_AWS_SECRETS__PREFIX | No | Prefix placed in front of every AWS secret reference. | |
| LIBRARIAN_AWS_SECRETS__REGION | No | AWS region for Secrets Manager. Default: the usual AWS configuration. | |
| LIBRARIAN_LEARNING__FIND_DOCS | No | Given only a base URL, look for documentation in the usual places. Default: true. | |
| LIBRARIAN_MAX_CALLS_PER_FETCH | No | The most API calls one plain-language request may make. Default: 8. | |
| LIBRARIAN_PRINCIPALS__ISSUERS | No | JSON list of identity provider issuer objects with issuer, audience, jwks_url, claim, algorithms and leeway_seconds. | |
| LIBRARIAN_SUBSTRATE__AUDIENCE | No | What a substrate token must be made for. Default: librarian. | |
| LIBRARIAN_SUBSTRATE__JWKS_URL | No | Where the issuer's keys are. Default: from its discovery document. | |
| LIBRARIAN_AUDIT_RETENTION_DAYS | No | How long audit events are also kept in the database for the console's Activity page. Default: 30. | |
| LIBRARIAN_DRIFT__RECHECK_HOURS | No | How often each API's spec or documentation is read again for changes; 0 never. Default: 24. | |
| LIBRARIAN_MCP_MAX_RESULT_BYTES | No | Largest MCP tool result execute sends, in bytes as sent. A larger one comes back as its outline; 0 means no limit. Default: 1000000. | |
| LIBRARIAN_MCP_MAX_WAIT_SECONDS | No | Longest write_status waits for a write to end when an agent asks with wait_seconds; 0 never waits. Default: 50. | |
| LIBRARIAN_PUBLISH__BASE_BRANCH | No | The branch written to. Default: main. | |
| LIBRARIAN_REGISTRY_OVERLAY_DIR | No | Without a database: where librarian writes its own entries when registry_dir is managed elsewhere. | |
| LIBRARIAN_FETCH_TIMEOUT_SECONDS | No | The time one plain-language request may take; past it, what it has is returned marked partial. Default: 100. | |
| LIBRARIAN_LEARNING__SEARCH__URL | No | The SearXNG instance used for documentation search. | |
| LIBRARIAN_NOTIFIER__WEBHOOK_URL | No | With notifier.type webhook: where each event is posted as JSON. | |
| LIBRARIAN_RATE_LIMIT_PER_MINUTE | No | Reads one agent may make on one API in a minute. Default: 60. | |
| LIBRARIAN_SUBSTRATE__ALGORITHMS | No | JSON list of signature algorithms accepted for substrate tokens. Default: ["ES256", "RS256"]. | |
| LIBRARIAN_SUBSTRATE__KAGENT_URL | No | kagent's controller. When set, the caller is the kagent Agent behind a session rather than the session. | |
| LIBRARIAN_WRITES__ANNOUNCE_AUTO | No | Post every automatically approved write to the channel. Default: false. | |
| LIBRARIAN_ALLOW_PRIVATE_NETWORKS | No | Let learned APIs reach private addresses. For a laptop; otherwise use internal_hosts. Default: false. | |
| LIBRARIAN_APPROVERS__USER_HEADER | No | The header in which the proxy names the person. Default: X-Auth-Request-Email. | |
| LIBRARIAN_WRITES__MAX_GRANT_DAYS | No | The longest a grant may last. Unset means any length, including until revoked. | |
| LIBRARIAN_CREDENTIAL_WAIT_SECONDS | No | How long an A2A task waits for a person to add a missing credential. Default: 900. | |
| LIBRARIAN_REQUEST_TIMEOUT_SECONDS | No | How long connecting to an API, or waiting for the next part of its answer, may take; one request as a whole may take twice this. Default: 30. | |
| LIBRARIAN_WRITES__ANNOUNCE_GRANTS | No | Post every use of a grant to the channel. Default: true. | |
| LIBRARIAN_APPROVERS__GROUPS_HEADER | No | The header in which the proxy lists the person's groups. Default: X-Auth-Request-Groups. | |
| LIBRARIAN_LEARNING__DOCS_MAX_PAGES | No | The most documentation pages read for one API. Default: 8. | |
| LIBRARIAN_NOTIFIER__SLACK__CHANNEL | No | The Slack channel; an API's entry can name its own. Default: #api-access-requests. | |
| LIBRARIAN_LEARNING__MODEL_KNOWLEDGE | No | Given only a base URL and no documentation, describe the API from what the model knows. Default: true. | |
| LIBRARIAN_LEARNING__SEARCH__API_KEY | No | The Brave Search key. | |
| LIBRARIAN_SUBSTRATE__LEEWAY_SECONDS | No | Clock difference tolerated when checking a substrate token's times. Default: 30. | |
| LIBRARIAN_AWS_SECRETS__CACHE_SECONDS | No | How long an AWS secret value is kept in memory. Default: 60. | |
| LIBRARIAN_LEARNING__SEARCH__PROVIDER | No | Search the web for an API's documentation: off, model, searxng or brave. Default: off. | |
| LIBRARIAN_NOTIFIER__SLACK__BOT_TOKEN | No | The Slack bot's token. | |
| LIBRARIAN_PUBLISH__GITHUB_APP__APP_ID | No | GitHub App id for the record repository. | |
| LIBRARIAN_NOTIFIER__EXPIRE_AFTER_HOURS | No | Close a credential request nobody answered after this long. Default: 24. | |
| LIBRARIAN_NOTIFIER__REMIND_AFTER_HOURS | No | Remind about a credential request still open after this long. Default: 4. | |
| LIBRARIAN_LEARNING__SEARCH__MAX_RESULTS | No | How many search results are looked at. Default: 8. | |
| LIBRARIAN_WRITES__APPROVAL_EXPIRE_HOURS | No | How long a write waits for a decision before it expires. Default: 24. | |
| LIBRARIAN_WRITES__RATE_LIMIT_PER_MINUTE | No | Writes one agent may ask for on one API in a minute; a plan counts as its number of steps. Default: 10. | |
| LIBRARIAN_SUBSTRATE__CLAIMED_USER_HEADER | No | Request header in which the agents' platform names the person at the agent, e.g. X-User-Id. It is only shown to approvers and recorded as what the agent's side says. | |
| LIBRARIAN_NOTIFIER__SLACK__APPROVER_GROUP | No | A Slack user group to mention. | |
| LIBRARIAN_SUBSTRATE__KAGENT_CACHE_SECONDS | No | How long the answer to which Agent a session is gets reused. Default: 60. | |
| LIBRARIAN_PUBLISH__GITHUB_APP__PRIVATE_KEY | No | The GitHub App private key, inline. | |
| LIBRARIAN_LEARNING__SEARCH__ALLOW_THIRD_PARTY | No | Also use pages outside the API's own domain and the well-known documentation hosts. Default: false. | |
| LIBRARIAN_PUBLISH__GITHUB_APP__INSTALLATION_ID | No | The GitHub App installation id. Found automatically when unset. | |
| LIBRARIAN_PUBLISH__GITHUB_APP__PRIVATE_KEY_FILE | No | A file holding the GitHub App private key. |
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
Server capabilities have not been inspected yet.
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
No tools | |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
This server cannot be deployed
Maintenance
ActivityMaintained
ResponsivenessNo recent community issues