> ## Documentation Index > Fetch the complete documentation index at: https://docs.murmur.dev/llms.txt > Use this file to discover all available pages before exploring further. # tool-link-policy catalog resource > Catalog resource that names a set of MCP tool-link extraction patterns — the rules that decide which MCP tool results are captured as agent artifacts. A tool link policy is a named list of extraction patterns selecting which MCP tool results become artifacts of an agent's execution — a Linear issue the agent filed, a Jira ticket, a Notion page — shown alongside its uploads and preview links in the dashboard. [Workspaces](/catalog/workspace) select a policy by name (`tool_link_policy`). The platform ships the builtin **`murmur-tool-links`**, which every workspace uses by default — currently Linear issue writes, labeled by the issue identifier (e.g. `MUR-123`). A workspace-selected policy **replaces** the builtin: copy its entries into your own policy to keep them, or select a policy with an empty `patterns` list to disable capture entirely. ## Fields | Name | Type | Required | Description | | ------------- | --------- | -------- | ------------------------------------------------------------------------------------------------------------------------------------------ | | `name` | string | yes | Unique identifier. DNS label format: `[a-z][a-z0-9-]{0,62}`. | | `description` | string | no | Human-readable description shown in the dashboard. Max 1024 bytes. | | `patterns` | object\[] | no | Extraction entries, evaluated independently against every MCP tool result. Empty = capture nothing. See [Pattern fields](#pattern-fields). | ## Pattern fields Each entry pairs a `tool` predicate with **exactly one extraction arm**: * **URL arm** (`url`) — capture URLs the tool result already contains. * **Identifier arm** (`id` + `link`) — construct the URL from an identifier in the result text, for servers whose results carry a key but no human URL (or only an unclickable API URL — e.g. Jira's create result returns the issue key and a REST self-link, not the browse page). | Name | Type | Required | Description | | ------- | ------ | --------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `tool` | string | yes | RE2 matched (unanchored) against the MCP tool name that produced the result. Claude names MCP tools `mcp____`; Codex names them `.`. A suffix pattern like `save_issue$` covers both backends. | | `url` | string | one arm | URL arm: RE2 matched (unanchored) against each canonicalized URL extracted from the tool result. Mutually exclusive with `id`/`link`. | | `id` | string | one arm | Identifier arm: RE2 with **one or more** capture groups, run against the tool result text. Group 1 is the identifier (the artifact's default label); further groups compose multi-variable URLs. A match constructs an artifact only when every group captured non-empty text. Requires `link`. | | `link` | string | with `id` | Identifier arm: URL template expanding `id`'s captures via `$1`..`$N` (write `${n}` when a word character follows). Must reference at least one capture and expand to a parseable `https` URL. | | `label` | string | no | RE2 with exactly one capture group, applied to the URL; a non-empty capture becomes the artifact's display label. Empty = the markdown link text (URL arm) or the captured identifier (identifier arm), then the URL host. | **Pick one arm per tool.** Entries evaluate independently, so configuring both arms for the same tool can record two artifacts for one creation (the result's own URL and the constructed one). Match on *write* tools (`save_issue$`, `create_issue$`), not read tools — the tool half of each entry is what keeps a 50-item listing result from becoming 50 artifacts. The same link re-surfacing later dedups to one artifact per execution. ## Examples ### Extending the builtin (copy, then add) ```yaml theme={null} name: team-links description: "Builtin Linear capture plus Notion pages and Jira tickets" patterns: # Copied from the builtin murmur-tool-links (a custom policy replaces it): - tool: save_issue$ url: ^https://linear\.app/[^/]+/issue/ label: /issue/([A-Za-z0-9]+-[0-9]+) # URL arm — Notion create results contain the page URL: - tool: create_page$ url: notion\.so/ # Identifier arm — Jira create results carry only the key; construct the browse URL: - tool: create_issue$ id: '"key"\s*:\s*"([A-Z]+-[0-9]+)"' link: https://acme.atlassian.net/browse/$1 # Multi-variable identifier arm — a Slack message permalink needs the channel # AND the ts (with its dot removed, expressed by splitting ts into two groups): - tool: post_message$ id: '"channel":"(C[A-Z0-9]+)".*?"ts":"([0-9]+)\.([0-9]+)"' link: https://acme.slack.com/archives/$1/p$2$3 ``` ```bash theme={null} cat team-links.yaml | murmur set tool-link-policy team-links ``` Then point the workspace at it: ```yaml theme={null} # in the workspace resource tool_link_policy: team-links ``` ### Disabling capture ```bash theme={null} echo 'name: no-capture' | murmur set tool-link-policy no-capture ``` A policy with no `patterns` captures nothing; set `tool_link_policy: no-capture` on the workspace. ### Inspecting the builtin ```bash theme={null} murmur get tool-link-policy murmur-tool-links ``` ## Errors | Code | Meaning | What to do | | --------------------- | ------------------------------------------------------------------------ | -------------------------------------------------------------------------------------- | | `INVALID_ARGUMENT` | `patterns[].tool: ...` / `.url: ...` | The pattern is empty or not valid RE2. Fix the regexp. | | `INVALID_ARGUMENT` | `patterns[].label: ... must have exactly one capture group` | The `label` regexp must contain exactly one `(...)` group. | | `INVALID_ARGUMENT` | `patterns[].id: ... must have at least one capture group` | The `id` regexp needs at least one `(...)` group — group 1 is the identifier. | | `INVALID_ARGUMENT` | `patterns[]: url and id/link are mutually exclusive arms` | The entry sets both extraction arms. Pick one: `url`, or `id` + `link`. | | `INVALID_ARGUMENT` | `patterns[]: url or id+link is required` | The entry has a `tool` but no extraction arm. | | `INVALID_ARGUMENT` | `patterns[].link: ... must reference an id capture` | The `link` template never uses `$1`..`$N`. Write `${n}` when a word character follows. | | `INVALID_ARGUMENT` | `patterns[].link: ... must expand to an https URL` | The expanded template is not a parseable `https` URL with a host. | | `INVALID_ARGUMENT` | `name "murmur-..." is reserved` | Names starting with `murmur-` are platform builtins. Pick another name. | | `FAILED_PRECONDITION` | `cannot delete tool-link-policy "": referenced by workspace: ` | A [workspace](/catalog/workspace) references this policy. Point it elsewhere first. | ## Related * [Workspaces](/catalog/workspace) — select a policy via `tool_link_policy` * [`murmur set`](/cli/set) / [`murmur get`](/cli/get) — create, update, and inspect policies