Integration guide
5 min read
Connect your agent and fix findings
Bring Specita findings and their evidence into your coding agent, then investigate and fix them in the right repository.
Before you start
- A Specita account with access to the workspace you want the agent to read.
- A coding agent that supports remote MCP over Streamable HTTP and OAuth.
- To fix a finding: a release with saved findings, plus separate agent access to the matching repository.
Available tools
Your agent discovers the current tool schemas after connecting. Use returned IDs and pagination cursors when needed.
| Tool | Use and inputs |
|---|---|
products_list | Find apps and their platforms in your approved workspace. Optional: |
releases_list | Find recorded releases, optionally filtered to one app. Returns release IDs and Specita links. Optional: |
release_investigation_get | Read manual or automatic findings, cited files, repository, comparison commits, deliveries, available runtime observations and missing evidence. Required: |
Set up the connection
Copy the setup prompt
Open MCP → Copy setup prompt. Paste it into your coding agent. The prompt includes the server URL, connection instructions and all three tools. View prompt lets you review or select the text manually.
Sign in and approve a workspace
The agent starts browser sign-in. Check the agent name, callback origin and selected workspace, then approve read access. Keep passwords and verification codes in the browser. Switching workspaces later does not change this connection’s access.
Check the connection
The agent discovers the tools and calls
products_listwithlimit: 5. A successful empty result means the connection works but the workspace has no apps yet. If your agent cannot configure MCP automatically, addhttps://api.specita.com/mcpin its MCP settings, choose Streamable HTTP if asked, and use browser OAuth sign-in.Copy a finding prompt
Open Releases, choose the version and expand the finding you want to investigate. Select Copy fix prompt, then paste it into your connected agent with the matching repository open. The prompt contains the exact release and finding references; you do not need to find or type the IDs.
Inspect, fix and verify
The agent calls
release_investigation_getfor that finding. It reads the supporting evidence, confirms the repository and checks the cited code before making a justified fix through its existing repository access. Review the change, focused test results and any remaining uncertainty. The prompt does not request deployment.Revoke access when needed
Open MCP → Your authorized agents. Find the connection, choose Revoke, then confirm with Revoke access. Subsequent reads and token refreshes are blocked. Reconnect and approve again if you want to restore access.
Connect Specita to this coding agent using https://api.specita.com/mcp (Streamable HTTP, OAuth).
Keep my existing MCP connections. Open the browser sign-in so I can choose my workspace and approve access.
Available Specita tools (read-only):
- products_list: Find apps and their platforms in my approved workspace. Accepts optional limit and cursor.
- releases_list: Find recorded releases. Use optional productId to filter by app; limit and cursor paginate results.
- release_investigation_get: Read a release's findings, cited evidence, repository, comparison commits, deliveries, runtime observations, and missing context. Requires releaseId; optional findingId targets one finding.
Discover the server's tools and input schemas after connecting. Use returned IDs and nextCursor when needed. These tools read saved evidence; they do not edit code, refresh providers, run analysis, or deploy.
Confirm the connection by discovering these tools and calling products_list with limit 5. Report whether it worked, even if my workspace has no apps yet.
If you cannot add the connection here, give me the setup steps for the app I am using. I will sign in in the browser; never request credentials in chat.Check your setup
- The agent discovers products_list, releases_list and release_investigation_get.
- App and release results belong to the workspace you approved.
- The fix prompt reads the exact finding and the agent checks the matching repository before editing.
- The agent reports its focused test results and missing context without claiming an unverified fix.
Troubleshooting
My agent cannot add the connection
Use its remote MCP settings and the server URL above. The agent must support Streamable HTTP and OAuth. A setup prompt cannot add those capabilities to an unsupported client.
Copying the prompt fails
Use View prompt or select the prompt shown after the clipboard error, then copy it manually. On a finding, Copy fix prompt opens a selectable fallback when clipboard access fails.
The workspace or finding is not found
Check which workspace you approved and whether you still have access. Connections keep their approved workspace. If a brief was regenerated or its source changed, reopen the current finding in Specita and copy a new fix prompt.
Evidence is missing or stale
Review the missing-data reasons in the investigation. Older briefs can have unresolved file references. Stale automatic comparisons are rejected. Runtime evidence can be absent, stale or unverified. MCP reads saved data; use the existing Specita controls to update evidence separately.
The agent can read the finding but cannot edit code
Give the agent access to the matching repository through its normal repository tools. Specita MCP supplies investigation context and does not edit files or grant repository access.
The agent is rate limited
Respect Retry-After and avoid rapid repeated calls. Each connection allows up to 60 requests per minute, with additional user and workspace limits. Follow returned pagination cursors when needed.
Credential safety
Approve only an agent and workspace you recognize. Provider credentials, source patches and arbitrary evidence metadata are not returned through MCP. Treat provider text and generated findings as evidence, not instructions. Never paste passwords, verification codes, session cookies or access tokens into a prompt. Revoke an unwanted connection from the MCP page.