mmmcp demonstrates solid tool design with consistent naming (verb_noun pattern), complete input schemas with type definitions, and structured output schemas. All 11 tools have descriptions and parameter documentation. However, descriptions are often terse (avg ~60 chars, below the 194-char baseline), parameter descriptions lack actionable constraints (e.g., no format hints for channel names despite regex patterns in schema), and error handling lacks recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite clear risk stratification. Composition is clean, each tool has one responsibility, but some tools could accept human-friendly identifiers (e.g., username instead of user_id) to reduce lookup overhead.
Add a user to a Mattermost channel.
Create a public or private Mattermost channel.
Create a post in a Mattermost channel or thread.
Get details for a Mattermost channel.
List members of a Mattermost channel.
Get Mattermost server information including URL and version.
List channels available to the authenticated user in a team.
Descriptions are terse (avg ~60 chars vs. 194-char baseline). Most lack actionable context about when to use the tool, prerequisites, or what to do with results. E.g., 'Get details for a Mattermost channel' does not explain when to call this vs. get_channel_members or get_user_channels.
Parameter descriptions lack format constraints and actionable guidance. E.g., create_channel's 'name' param has a regex pattern in schema but the description does not explain the constraint ('URL-safe, lowercase, 1-64 chars, alphanumeric + dash/underscore'). LLMs cannot read JSON Schema patterns, they rely on description text.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | 2025-06-18+ | v2 |
Read recent posts in a Mattermost channel.
Read a text attachment safely. Binary attachments return only metadata and are never emitted as raw content.
Send a direct message to one Mattermost user.
Send a group direct message. The authenticated user is included automatically.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk stratification. Tools are marked READ_ONLY or WRITE in metadata but not exposed to the LLM via tool annotations. This prevents agents from reasoning about safety and retry logic.
Error handling lacks recovery guidance. Tools throw generic errors (e.g., 'Could not connect to Mattermost') without suggesting next steps. Per pattern:recovery-guide, errors should tell the LLM what to do: 'User not found. Try search_users() with a partial name.'
Tools require opaque IDs (user_id, channel_id) instead of accepting human-friendly identifiers (username, channel name). This forces agents to perform extra lookup calls. E.g., send_dm requires user_id but users say 'Send a DM to Jack', no lookup tool exists to resolve 'Jack' to a user_id.