Platform MCP
The platform MCP lets an agent (Claude Code, Cursor, VS Code, Codex, or your own agent) do what you can do in the dashboard: run and triage analyses, manage rules, products and repositories, administer your organization and, for a platform admin, the platform. It acts as you, never with more rights.
It is the only MCP endpoint. A CI job connects with an org API key that has the pipeline role:
it sees the analysis tools and nothing else.
Connect
Section titled “Connect”| Where | URL |
|---|---|
| Its own host | https://mcp.<your-domain>/ |
| On the API host | https://api.<your-domain>/platform-mcp/ |
| Local stack | http://localhost:8080/platform-mcp/ |
In the dashboard: Account → Agent access. Create an access token for yourself, then:
claude mcp add --transport http opentremor https://mcp.example.com/ \ --header "Authorization: Bearer ot_pat_..."Any MCP client that takes a JSON server list:
{ "mcpServers": { "opentremor": { "url": "https://mcp.example.com/", "headers": { "Authorization": "Bearer ot_pat_..." } } }}An org API key (X-API-Key) works too, for CI or a shared bot that acts in one organization:
claude mcp add --transport http opentremor https://mcp.example.com/ \ --header "X-API-Key: $OPENTREMOR_API_KEY"Hosted clients (OAuth)
Section titled “Hosted clients (OAuth)”Claude.ai and ChatGPT connectors take only the URL. Add https://mcp.<your-domain>/ as a custom
connector: the client opens an OpenTremor page where you sign in, choose the scope and, optionally,
one organization, and allow it. Disconnect it in the dashboard under Account → Agent access → Connected apps.
See Access Tokens for Agents.
What the agent sees
Section titled “What the agent sees”The agent sees only the tools your credential allows: an access token’s scope and your role in the
organization decide the list. A read token gets read tools only; the platform tools appear only
for a platform admin’s admin token.
Tools that take org_id act in the organization you specify. If you omit it, they use the
organization in the caller’s current context. For access tokens, you can pass org_id to act in
another organization you belong to. The MCP server forwards an explicit org_id to the API as
X-OpenTremor-Org.
Tools marked destructive take confirm: true. Without it they describe what they would do, so the
agent can ask you first.
Left out on purpose
Section titled “Left out on purpose”Some actions stay in the dashboard: writing SSO, SCIM or sign-in methods (a mistake locks everyone out), deleting an organization, granting platform admin, and anything about your own account (password, 2FA, tokens).
Analyse a change yourself
Section titled “Analyse a change yourself”The agent runs the client-led loop with its own model:
ingestsplits the input into units and returns a receipt with therevision_id.claim_unitshands out a batch of units (5 by default, 20 at most) with their rules and thecontractthat says how to write an analysis. Itsstatussays what to do next:units(analyse and submit them),waiting(another caller holds the remaining units; claim again afterretry_after_seconds) ordone.submit_analysesstores the analyses of several units in one call and reports each unit’s result:stored,rejectedwith errors to fix,not_in_revision, orclaimed_by_other.get_reportreturns the report once the status isdone.
Each ruleset and the contract are sent once: a unit names its ruleset_id, and the agent lists
the ids it already holds in known_rulesets and known_contract. An agent that stops early calls
release_units, so the units it claimed do not wait out their 10-minute claim.
The server also offers:
| Kind | Name | What it is |
|---|---|---|
| Prompt | analyse_changeset | The loop above, as instructions for the agent |
| Resource | opentremor://packs/{pack}/rules | A pack’s ruleset in your current organization |
| Tool | Kind | Needs | What it does |
|---|---|---|---|
whoami | read | — | Who you are acting as: user, organization, role, permissions, and whether through an access token |
list_my_orgs | read | — | The organizations you belong to, with your role in each, under items |
list_packs | read | analysis:read | The packs enabled for the organization: the kinds of input you can analyse |
get_pack_rules | read | analysis:read | A pack’s ruleset in Markdown, under text, to preview it |
ingest | write | analysis:run | Split raw input (a plan, a diff) into units under changeset |
claim_units | write | analysis:run | Claim up to limit units (1 to 20) of changeset to analyse |
submit_analyses | write | analysis:run | Store the analyses of claimed units, one entry per unit, in one call |
release_units | write | analysis:run | Hand back the units you claimed in changeset and will not submit, so they are claimable at once |
start_server_analysis | write | analysis:run | Let the server call the LLM and analyse raw_input in one step, with one of the org’s LLM connections |
list_changesets | read | analysis:read | The organization’s changesets, newest first, each with its repository, that repository’s product, its pull request and its open findings by severity |
list_changeset_revisions | read | analysis:read | A changeset’s revisions, newest first: one per push or analysis |
get_revision_changes | read | analysis:read | The findings one revision opened, closed and left open, compared with the revision before it |
get_report | read | analysis:read | The report of a changeset: its current revision, or revision_id |
get_merge_gate | read | analysis:read | Whether a changeset’s findings allow its change to merge, and why |
list_findings | read | findings:read | Findings for the organization, filtered |
findings_summary | read | findings:summary | Counts of the organization’s findings by status and severity |
triage_finding | write | findings:triage | Set a finding’s triage status, with a note explaining why |
submit_feedback | write | feedback:write | Record a verdict on a finding |
list_rules | read | rules:read | The organization’s custom rules |
get_rule | read | rules:read | One custom rule |
get_rule_coverage | read | rules:read | Where a custom rule runs: for each repository, runs or the first reason it does not. The reasons, in the order they are checked: rule_disabled, pack_off, other_variant, out_of_scope, excluded |
reset_rule | write | rules:manage | Restore a built-in rule’s pack wording, as shown by pack_update on a rule the organization edited |
create_rule | write | rules:manage | Add a custom rule |
update_rule | write | rules:manage | Change a custom rule |
delete_rule | destructive | rules:manage | Delete a custom rule |
list_rule_categories | read | rule_categories:read | The organization’s rule categories, to file a custom rule under |
list_rule_proposals | read | feedback:read | Rule changes suggested by review feedback or by the organization’s ADRs |
review_rule_proposal | write | rule_proposals:manage | Apply, defer or dismiss a rule proposal |
rule_health | read | feedback:read | How often reviewers agree with each rule: which rules are noisy |
list_products | read | products:read | The organization’s products: groups of repositories owned by teams |
get_product_overview | read | findings:summary | A product’s repositories, each with its open findings by severity, its number of changesets and its last revision, and the product’s totals |
get_product_report | read | findings:read | A product’s open findings: one section per repository with its recent changesets, after the rules with the most findings |
get_repository_report | read | findings:read | A repository’s open findings and recent changesets |
create_product | write | products:manage | Add a product, optionally owned by some teams and holding some repositories |
list_goals | read | products:read | Where a product is heading: its goals, each with ordered steps and their status |
create_goal | write | goals:manage | Add a goal to a product, with its steps in order |
get_goal | read | products:read | One goal with its steps, and the pull requests that advanced or regressed it, newest first |
update_goal_step | write | goals:manage | Set a goal step’s status: todo, in_progress, done or skipped |
list_repositories | read | repositories:read | The organization’s repositories known to OpenTremor, filtered |
update_repository | write | repositories:manage | Change a repository’s settings |
resolve_pack_suggestion | write | repositories:manage | Act on a pack in a repository’s suggested_packs |
preview_repository_scan | read | repositories:scan | What a scan of a repository’s existing code would read |
start_repository_scan | write | repositories:scan | Scan a repository’s existing code on its default branch |
get_repository_scan | read | repositories:read | A repository’s last scan: its status, the slices it ran, those left to resume, its job, and the units it analysed and reused |
list_org_packs | read | packs:manage | Which packs are enabled for the whole organization |
set_pack_enabled | write | packs:manage | Enable or disable a pack for the whole organization |
list_jobs | read | jobs:list | The organization’s analysis jobs, newest first |
get_job | read | jobs:read | One job’s status and progress |
cancel_job | write | jobs:run | Stop a queued or running job |
get_org | read | org:read | The organization’s details |
list_members | read | members:read | The organization’s members and their roles |
change_member_role | write | members:manage | Give a member another role |
remove_member | destructive | members:manage | Remove a member from the organization |
invite_member | write | invites:manage | Invite someone by email |
list_invites | read | invites:manage | The organization’s open invitations |
list_teams | read | teams:read | The organization’s teams |
create_team | write | teams:manage | Add a team |
add_team_member | write | teams:manage | Add a member of the organization to a team |
list_roles | read | roles:read | Built-in and custom roles, with the permissions each holds |
list_api_keys | read | api_keys:manage | The organization’s API keys (service accounts), without their secrets |
create_api_key | write | api_keys:manage | Create an API key for a CI job or a shared bot |
revoke_api_key | destructive | api_keys:manage | Revoke an API key |
get_agent_access | read | agent_access:manage | Whether members’ access tokens act in the organization |
set_agent_access | write | agent_access:manage | Allow members’ access tokens, cap them at read, or refuse them in the organization |
get_sso | read | sso:manage | The organization’s SSO connection, read-only: changing it stays in the dashboard |
list_notification_channels | read | notifications:manage | Where the organization’s notifications go |
list_llm_connections | read | llm_connections:read | The organization’s LLM connections, for start_server_analysis |
get_usage | read | billing:read | The organization’s usage and spend this month |
query_audit_log | read | audit:read | The organization’s audit log, newest first |
admin_list_users | read | platform admin | Every account on the platform, filtered by name, email or organization |
admin_set_user_active | write | platform admin | Activate or deactivate an account across the platform |
admin_list_orgs | read | platform admin | Every organization on the platform |
admin_get_org | read | platform admin | One organization, as seen by a platform admin |
admin_set_org_plan | write | platform admin | Set an organization’s plan (cloud deployments only) |
admin_lock_org | write | platform admin | Lock or unlock an organization |
admin_get_settings | read | platform admin | The platform settings: each field’s value and whether it is overridden |
admin_update_settings | write | platform admin | Change platform settings |
admin_query_audit | read | platform admin | The platform-wide audit log, newest first |
Running it
Section titled “Running it”The service is OpenTremor-mcp. It is stateless, holds no secret, and calls the API with each
caller’s own credential.
| Variable | Default | Meaning |
|---|---|---|
OPENTREMOR_API_URL | http://localhost:8000 | Where the API is, as this service reaches it |
PORT | 8090 | Port of the MCP endpoint |
MCP_PATH | / | Path of the MCP endpoint |
MCP_PUBLIC_URL | — | This service’s public URL. Together with the next one, it enables OAuth discovery |
OPENTREMOR_PUBLIC_API_URL | — | The API’s public URL: the OAuth authorization server |
GET /healthz answers without a credential. A request to the MCP endpoint without a credential gets
401. See OpenTremor-mcp#0001 for the design.